Stop needless copying of inputs #10
Conversation
|
The import in metamouse.wdl should be pointing to this branch to take up the changes on this branch, of course, but it already passes/fails when expected on Terra so I went ahead and reverted it to point at main again so I won't forget to do that after meging. |
arraycheck_rdata.wdl
Outdated
| for i in ~{sep=' ' truth} | ||
| do | ||
| truth_basename="$(basename -- ${i})" | ||
|
|
||
| # We assume the test file and truth file have the same basename | ||
| # Due to how WDL input work, they have a different absolute path |
There was a problem hiding this comment.
This sort of thing keeps happening. I code in Sublime Text, which can spell check in comments, but it doesn't seem to recognize WDL comments as comments ("toggle comment" doesn't work even with the Broad's WDL syntax highlighter active and it seems most people just have issues with the keystroke but I'm using the menu). Still, there is a grammar check package out there, so if I can get it recognize comment scope, I should be able to stop this from happening...
Before I go down the force-sublime-to-recognize-a-comment rabbithole, do you know if there's something like lint for spelling/grammar comments? That might be easier than trying to wrangle Sublime. I've had issues with Package Control anyway.
There was a problem hiding this comment.
Only the blinding mental flare of the English language wrought into a shape that was never meant to exist.
This task takes in an array of truth files and an array of test files. Truth files are assumed to match the basename of test files. When a truth file is found to match a test file in basename, they are md5'd. If that fails, an Rscript checks if they pass
all.equal()within a user-definedtolerancewhich defaults to 1.0e-8. It should go without saying that the Rscript isn't going to function correctly if passed a non-rdata file.In a typical checker workflow, this task would be used to test multiple rdata outputs of a workflow's task against multiple truth files from a Google cloud bucket. But for the purposes of testing, both truth and test files are being downloaded from the cloud.
The testing confuguration is metamouse.wdl, which runs this workflow three times. In one of those tests, it is designed to pass. In the other two, it is designed to fail -- once because the
exactflag istrue, causing the file that that doesn't perfectly match to trigger a failure, and once because the deviation between two files is outside the value oftolerance. These are functionally equivalent but algorithmically different as the former does not call an rscript but the latter does.